Income from the account beats income from a document, and the reason is not fraud

The usual case for account-based income verification is that documents can be forged. True, and it undersells the argument.

Even a completely authentic pay stub is a poor instrument for the decision you are making, because it describes two weeks that the applicant chose to show you.

Four things a stub cannot tell you

Duration. Two stubs prove the applicant was employed for four weeks. They say nothing about whether that job is three weeks old. Twelve months of deposits says how long the income has existed, which is closer to the question of whether it will continue.

Variance. Two consecutive stubs from a variable earner tell you about two periods. An applicant averaging $4,200 with a range of $3,900 to $4,500 and one averaging $4,200 with a range of $1,100 to $8,000 look identical on two stubs and are different residents.

Every source. An applicant with W-2 wages, a weekend gig, and rental income from a property has three streams. A stub shows one. The account shows all three without three document requests, and it shows the ones the applicant forgot to mention.

Direction. Deposits trending down over six months is a signal no stub carries. It is also one of the more predictive things available about the next twelve months.

Those four are why this is a measurement argument rather than a fraud argument. Fraud detection is what you need when the document is the only artifact available. Account connection means the document is not the artifact.

What connecting actually involves for the applicant

The friction concern is the main objection and it is worth being concrete.

The applicant opens a link in any browser. No app to install, no account to create. They select their bank or payroll provider, authenticate with that institution's own credentials, and grant read access. ProofUp uses Plaid for this, which is the same mechanism most consumer financial apps use, so a meaningful share of applicants have done it before.

Two things about it that matter operationally.

Credentials go to the institution rather than to the property or to us. That is the part worth telling applicants, because the reasonable version of their hesitation is "why would I give my bank password to a leasing office," and the answer is that they are not.

It takes about the same time as photographing and uploading two pay stubs, and it produces a verdict in two to three minutes rather than whenever somebody reviews the upload.

The averaging decision matters more than the data source

Having twelve months of deposits raises a question two stubs never forced you to answer: what number do you use.

The options and what they do:

Mean over twelve months. Simple, defensible, and it understates risk for a high-variance earner because it treats a $900 month and a $7,000 month as averaging fine.

Median. Less sensitive to one unusual month in either direction. A reasonable default.

A low percentile, for example the 25th. Conservative and honest for variable income, since it asks whether a bad-but-normal month covers rent.

Whatever you pick, the requirement is uniformity. The same window and the same statistic for every applicant, written down. A statistic chosen per applicant is a discretionary judgment about income, applied inconsistently, which is the shape of a fair housing finding regardless of intent.

What deposits do not distinguish on their own

Being straight about the limits, because deposit analysis is not magic.

A deposit is money arriving. It does not inherently know whether it is wages, a transfer from savings, a loan from a relative, a tax refund, or a one-time sale. Payroll deposits are identifiable by pattern and by descriptor, and irregular deposits require interpretation.

Which means an account connection that returns a raw deposit total is worse than a stub, and one that classifies payroll income specifically is better. The classification is the product, not the connection.

Payroll-provider connection is stronger than bank connection for this reason: it returns employer, pay frequency, gross and net, and year-to-date figures from the payroll system itself, with no classification needed. Where an applicant's payroll provider is connectable, that is the best available artifact.

What changes in the outcomes

Two numbers from our own portfolio, with what they measure.

50% less delinquency, measured against customers' prior process.

24% fewer evictions.

The mechanism is not stricter approval. It is that the approve-or-decline line moves to a real income figure, so some applicants who passed on a document now fail, and some who had no acceptable document now pass. The composition of the approved pool changes more than its size.

The related number that explains where the risk was concentrated: applicants approved by override default at roughly 4 times the rate of applicants approved on the standard. That is not an argument against overrides. It is an argument for counting them, which most operations do not.

Two things we do not do

We do not verify income that never reaches an account. Cash paid and never deposited is invisible to every method, including this one.

We do not set your qualification threshold. What multiple to hold, which statistic to average on, and how to treat variable income are policy decisions that belong to you and need to be written down.

Check one file both ways

Take an approved application with pay stubs. Ask that resident, if they are still with you, or use the next applicant who connects.

Compare the stub-derived monthly figure against the twelve-month deposit picture. Look specifically at the range, not the average.

The average will usually agree. The range is what you did not know.

How wide was it?